Weird issue

[Macker]

hey guys,

I'm currently working on a project (max 2017/VRay 3.4) that has got a strange problem.

I'm using the progressive renderer to produce images and it's been working perfectly up until a certain point when all CPU usage seems to drop off. On render nodes it drops to 0% and idles there with the job still loaded in memory, and on my local machine it does this:



Restarting the spawners on the nodes seems to fix the issue and they start running again at 100%, but eventually the same thing happens. I haven't found a way of getting my local machine to not do this, short of cancelling the render (which is not ideal when it's been running for an hour or two). Neither my local machine, nor the render nodes appear to have anything unusual in their logs - everything appears (in max/vray) as if the machines are all rendering just fine, however this appears not to be the case in the task manager.

I have no idea how long it takes for the machines to reach this point - I have only just noticed it.

I'm reluctant to upgrade to the latest version of VRay at the moment because i'm in the middle of a job - but will do once this is out the door.

[Macker]

I just got the render node going again and waited (5 minutes - check the time stamps) for it to stop working.

[2017/Feb/28|16:39:39] Setting up 56 thread(s)
[2017/Feb/28|16:41:50] Threads completed
[2017/Feb/28|16:41:50] 40 % completed.
[2017/Feb/28|16:41:50] 50 % completed.
[2017/Feb/28|16:41:50] 60 % completed.
[2017/Feb/28|16:41:50] 70 % completed.
[2017/Feb/28|16:41:50] 80 % completed.
[2017/Feb/28|16:41:50] 90 % completed.
[2017/Feb/28|16:41:50] Compiling lights geometry.
[2017/Feb/28|16:41:50] Preparing ray server.
[2017/Feb/28|16:41:50] Using embree ray tracing.
[2017/Feb/28|16:42:11] Building embree static trees took 16087 milliseconds
[2017/Feb/28|16:42:11] QND tree contains 2 primitives.
[2017/Feb/28|16:42:11] QND tree takes 1.00 MB (depth 2). Build time 5.88 ms.
[2017/Feb/28|16:42:11] Scene bounding box is [-1609.68,-30,-726.315]-[2027.41,2311.92,1273.68]
[2017/Feb/28|16:42:11] Preparing direct light manager.
[2017/Feb/28|16:42:11] 12 render channels in sequence data
[2017/Feb/28|16:42:11] Preparing global light manager.
[2017/Feb/28|16:42:11] Irradiance sample size is 84 bytes
[2017/Feb/28|16:42:11] Cannot use fast GI because secondary engine is not brute force or none
[2017/Feb/28|16:42:11] Broadcasting NOTIFY_PRE_RENDERFRAME.
[2017/Feb/28|16:42:11] beginFrame() done.
[2017/Feb/28|16:42:11] Broadcasting NOTIFY_BEGIN_RENDERING_ACTUAL_FRAME.
[2017/Feb/28|16:42:11] [DR] Calling renderImage()
[2017/Feb/28|16:42:11] Photon size is 56 bytes.
[2017/Feb/28|16:42:11] Light cache sample size is 100 bytes.
[2017/Feb/28|16:42:11] Building light cache.
[2017/Feb/28|16:42:11] Tracing 1000000 image samples for light cache in 64 passes.

[2017/Feb/28|16:42:11] Setting up 56 thread(s)
[2017/Feb/28|16:42:11] Threads completed
[2017/Feb/28|16:42:12] Light cache contains 362609 samples.
[2017/Feb/28|16:42:12] Light cache samples collected; compiling lookup tree.
[2017/Feb/28|16:42:12] Light cache takes 55.2 MB.
[2017/Feb/28|16:42:12] Prefiltering light cache.
[2017/Feb/28|16:42:12] 10 % completed.
[2017/Feb/28|16:42:12] 20 % completed.
[2017/Feb/28|16:42:12] 30 % completed.
[2017/Feb/28|16:42:12] 40 % completed.
[2017/Feb/28|16:42:12] 50 % completed.
[2017/Feb/28|16:42:12] 60 % completed.
[2017/Feb/28|16:42:12] 70 % completed.
[2017/Feb/28|16:42:12] 80 % completed.
[2017/Feb/28|16:42:12] 90 % completed.
[2017/Feb/28|16:42:12] Average rays per light cache sample: 22.43 (min 1, max 2549)
[2017/Feb/28|16:42:12] Using premultiplied light cache.
[2017/Feb/28|16:42:12] Rendering image.
[2017/Feb/28|16:42:12] Setting up 56 thread(s)
[2017/Feb/28|16:42:12] Start progressive rendering, server mode 1, perform rendering 1

It's at this point that all processing drops off to 0%

[yavor.rubenov]

Hello,

When the CPU usage drops - is the number of passes in the progress still changing or it stays at 312? Does it happen if you render only with the local machine and no DR?

Is it possible that any of the DR machines dropped the rendering at some point (crashing, power outage or something like this)? That would be a problem with the progressive sampler - it currently doesn't handle node disconnects and if that happens it will wait for the disconnected node forever.

Also - is it scene dependent and if so - could you share the problem scene ?

Best regards,
Yavor